HAL Id: hal-01312603
https://hal.archives-ouvertes.fr/hal-01312603
Preprint submitted on 7 May 2016
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires
Diophantine approximation with improvement of the simultaneous control of the error and of the denominator
Abdelmadjid Boudaoud
To cite this version:
Abdelmadjid Boudaoud. Diophantine approximation with improvement of the simultaneous control of the error and of the denominator. 2016. �hal-01312603�
Diophantine approximation with
improvement of the simultaneous control of the error and of the denominator
Abdelmadjid BOUDAOUD Department of Mathematics,
Faculty of Mathematics and Computer Sciences, University of M’sila, Algeria
Laboratory of Pure and Applied Mathematics (L.M.P.A.) May 7, 2016
Abstract
In this work we proof the following theorem which is, in addition to some other lemmas, our main result:
theorem. Let X =f(x1,t1), (x2,t2), ..., (xn,tn)g be a …nite part of R R +, then there exist a …nite part R of R + such that for all
" >0there existsr 2Rsuch that if0< " rthen there exist rational numbers pi
q i=1;2;:::;n
such that:
8<
:
xi pi
q "ti
"q ti
,i= 1;2; :::; n. (*) It is clear that the condition "q ti fori= 1;2; :::; n is equivalent to
"q t = M in
i=1;2;:::;n (ti). Also, we have (*) for all " verifying 0 < "
"0= minR.
The previous theorem is the classical equivalent of the following one which is formulated in the context of the nonstandard analysis ([2],[5],[6],[8]).
theorem. For every positive in…nitesimal real ", there exists an un-
( x = px
q +"
"q = 0
.
For this reason, to prove the nonstandard version of the main result and to get its classical version we place ourselves in the context of the nonstandard analysis.
1991 Mathematics Subject Classi…cation. 11J13, 03H05, 26E35.
Key words and phrases. Diophantine approximation, Farey series, Non- standard Analysis.
1 Introduction, Notations and Rappel
We dispose in the domain of Diophantine approximation of many results (refer for example to [3], [7]). In the following, we give as an example, the two most used theorems:
Theorem (Dirichlet) 1.1. [7]. Suppose that x1; x2; ::: ;xn are n real numbers and that T >1is an integer. Then there exist integers q,p1,p2,:::,pn
with
8<
:
xi pi q
1
T q (i= 1;2; :::; n) 1 q < Tn
. (1.1)
Theorem (Kronecker) 1.2. [7]. For any reals 1; 2; ::: ; nand anyt >0, the system of inequalities
8>
><
>>
:
jq 1 p1 1j< t jq 2 p2 2j< t
::::::::::::::::::
jq n pn nj< t
(1.2)
is solvable in integers q, p1; p2; ::: ;pn if and only if 1; 2; ::: ; n are not rationally dependent. Note that 1; 2; ::: ; n are said rationally dependent if there exist integers r, r1; r2; ::: ;rn not all zero such that
r1 1 +r2 2+:::+rn n =r.
When we take 1 = 2 =::: = n= 0, this theorem is used to approximate the reals i by using rationals pi
q to errors smaller than t q.
In general, in these results we observe that the simultaneous control be- tween the error and the common denominatorq should be clari…ed and spec- i…ed. This, because the approximation to a given error (which is generally small) requires a denominator that is generally too big. Conversely, the ap- proximation with a small denominator might give an error that is not really small. This question has motivated us to give the following theorem which is, in addition to some other lemmas, our main result of this work.
Theorem 1.3. Let X = f(x1,t1), (x2, t2), ..., (xn, tn)g be a …nite part of R R +, then there exist a …nite part R of R + such that for all " > 0 there exists r 2 R such that if 0< " r then there exist rational numbers
pi
q i=1;2;:::;n
such that:
8<
:
xi pi
q "ti
"q ti
, i= 1;2; :::; n. (1.3) We note that in (1.3) the condition "q ti for i= 1;2; :::; n is equivalent to
"q t = M in
i=1;2;:::;n (ti). Also, under the assumption of theorem 1.3, for all "
verifying0< " "0 = minR we obtain (1.3).
The theorem 1.3 is the classical equivalent of the following theorem (the- orem 1.4.) formulated in the context of the nonstandard analysis.
Theorem 1.4. For every positive in…nitesimal real", there exists an integer Q depending only of ", such that8stx2R 9 Px 2Z:
8<
:
x = Px
Q +"
"Q = 0
. (1.4)
In the following we make a comparison between our result (theorem 1.3) and the existing results such as Dirichlet’s theorem and Kronecker’s theorem.
Our main result is used to approximate at a reduced common denominator q since "q t (i.e. q t
") and at a di¤erent errors since xi pi
q "ti for i= 1;2; :::; n. In addition, if we taket1 =t2 =:::=tn=t >0and "0 =M in R then for every 0< " " there exist integers q,p ,p ,:::,p such that
M ax
i2f1;2;:::;ng
xi pi
q "tand q t
" (1.5)
i.e., a denominatorq t
" enough for an error not exceeding "t.
Look when we use, under the same hypotheses, the Dirichlet’s theorem.
It may happen that when we take 1
T > "t, the common denominator q 1 is small enough so that the maximum error is strictly greater than "t i.e.
"t < M ax
i2f1;2;:::;ng xi pi q
1 T q
1
T. In contrast, when we take T satisfying 1
T "t then we are sure that the maximum error is smaller than or equal to"ti.e. M ax
i2f1;2;:::;ng xi pi q
1 T q
1
T "t. But in this case it may happen that the common denominator q, since that 1 q < Tn, is very close to
Tn 1
("t)n ( q = Tn 1 1
("t)n 1; for instance). Consequently, to be sure of the realization of the approximation asked, it is necessary to choose
1
T "tand q can be too big in this case as we have seen.
On his part the Kronecker’s theorem is purely existential and don’t say anything on the common denominator.
From the above we can see that the theorem 1.3 ensure the ability to control the size of q and of the maximum error; especially when " (resp. n) become small (resp. large). For its proof we place ourselves in the framework of the nonstandard analysis and we proceed as follows :
(1) We …rst show theorem 1.4 (In the sequel noted theorem 2.1.) by using some lemmas.
(2) We translate theorem 1.4 by using the Nelson’s algorithm.
1.1 Notations
i) For a number x (integer or non) we have the following usages:
1) Abbreviation, st(x) indicates thatx is standard; 8stx signi…es 8x[st(x) =)::].
2) x= +1 ( resp. x= 0) signi…es thatx is a positive unlimited (resp. x an in…nitesimal). x >
= 0 signi…es thatx is an in…nitesimal real strictly positive.
3) $(resp. ) signi…es a limited real (resp. an in…nitesimal real) on which one doesn’t say anything besides.
4) kxk is the di¤erence, taken positively, betweenx and the nearest integer.
5) E(x) (resp. fxg) is the integral part of x (resp. the fractional part ofx;
that is fxg=x E(x)).
6) Let " be an in…nitesimal real, one designates by " galaxie(x) the set fy : y=x+"$g and by " halo(x) the setfy : y=x+" g.
7) x0 signi…es, forx limited, the standard part of x.
ii)
8) IfE is a given set,E (resp. jEj) designates the external set formed, only, by the standard elements of E (resp. the cardinality of E).
9) One notes by (x1; x2; :::; xn)T the vector column 0 BB B@
x1 x2 ... xn
1 CC CA.
1.2 Rappel
1.2.1 Farey series([3])
The Farey seriesFN of orderN is the ascending series of irreducible fractions between 0 and 1 whose denominators do not exceed N. Thus h
k belongs to FN if
0 h k N,(h; k) = 1 the numbers 0and 1are included in the forms 0
1 and 1 1. If h
k < h0 k0 < h00
k00 are three successive elements of FN (N >1), then one has the following proper- ties:
10) kh0 hk0 = 1.
20) h0
k0 = h+h00 k+k00. 30) k+k0 > N and h
k < h+h0 k+k0 < h0
k0.
40) IfN >1, two successive elements ofFN don’t have the same denominator.
50) Let h1
k1, h2
k2 be two successive elements ofFN (N 1) with h1
k1 < h2
k2, and let the two following sequences:8
><
>:
U0 = h2
k2, U1 = h2+h1
k2+k1, ... ,Ui = h2+ih1 k2+ik1, ...
V0 = h1
k1, V1 = h1+h2
k1+k2, ..., Vj = h1+jh2 k1+jk2, ...
. (1.6)
We prove easily that the sequence(Ui)i2N (resp. (Vj)j2N) is decreasing (resp.
increasing); besides we have:
8>
<
>:
Ui Ui+1 = 1
(k2+ik1) (k2+ (i+ 1)k1), Ui h1
k1 = 1
k1(k2 +ik1)
Vj+1 Vj = 1
(k1+jk2) (k1+ (j + 1)k2), h2
k2 Vj = 1
k2(k1 +jk2) . (1.7) 1.2.2 Approximation to the in…nitesimal sense of reals
Theorem 1.5.[1]. Let be a real number. Then for all positive in…nitesimal real " there exist a rational number p
q and a limited real l such that:
( = pi
q +"l
"q = 0 . (1.8)
2 Simultaneous approximation to the in…ni- tesimal sense of standard reals
We prove in this section the following theorem whose translation by the al- gorithm of Nelson gives the theorem 1.3 .
Theorem 2.1. For every positive in…nitesimal real", there existsan integer Q depending only of ", such that 8stx2R9 Px2Z:
8<
:
x = Px
Q +"
"Q = 0
. (2.1)
Let" be a positive in…nitesimal real. We need to the following lemmas
Lemma 2.2. Let ( 1; 2; :::; N) a system of real numbers with N 1 limited. Then for all positive in…nitesimal real there are rational numbers
pi
q i=1;2;:::;N
and limited reals (li)i=1;2;:::;N such that fori= 1;2; :::; N : (
i = pi
q + li
q = 0
. (2.2)
Proof. Consider, for everyn2N , the formula:
B(n) =
"8 ( 1; 2; :::; n)2Rn with n 1and 8 >
=
09 Pi
Q i=1;2;:::;n
such that for everyi2 f1;2; :::; ng : 8<
:
xi Pi
Q = $
Q= 0
"
.
By theorem 1.5, we have B(1). Suppose, for 1 n a standard integer, B(n) and prove B(n+ 1). Let 1; 2; :::; n; n+1 2 Rn+1 and let >
= 0, then by B(n) there are rational numbers pi
q i=1;2;:::;n
such that 8>
>>
>>
><
>>
>>
>>
:
1 = p1
q + $
2 = p2
q + $
... = ...
n = pn
q + $
(2.3)
where q = 0. Now, since q = 0, the application of theorem 1.5 implies q n+1 = pn+1
qn+1 + ( q)$,( q)qn+1 = 0. Hence
n+1 = pn+1
qqn+1 + $, qqn+1 = 0. (2.4)
We deduct from (2.3) and (2.4) that:
8>
>>
>>
>>
>>
>>
><
>>
>>
>>
>>
>>
>>
:
1 = p1qn+1
qqn+1 + $ = P1
Q + $
2 = p2qn+1
qqn+1 + $ = P2
Q + $
... = ... = ...
n = pnqn+1
qqn+1
+ $ = Pn
Q + $
n+1 = pn+1
qqn+1 + $ = Pn+1
Q + $
where, from (2.4), Q= qqn+1 = 0. ConsequentlyB(n+ 1). Therefore, by the external recurrence principle, we have 8stn 1 B(n).
Lemma 2.3. LetE be a given set. For all integer! = +1, there is a …nite subset F E containing all standard elements of E (i.e. E F ) and whose cardinal is strictly inferior to ! (jFj< !).
Proof. Let! = +1. LetB(F; z)be the internal formula: "F E,jFj< !, z 2 F ". Let Z E be a standard …nite part. Then there exists a …nite part F E with jFj < ! such that every element z of Z belongs to F, i.e.
we have B(F; z). Indeed it su¢ ces to take F =Z. Therefore, the principle of idealization (I) asserts the existence of a …nite part F E with jFj< ! such that any standard element of L belongs toF.
Lemma 2.4. Let = +1be a real number such that p
" = 0. LetFM be the Farey sequence of orderM =E p
" . If p1 q1, p2
q2 are two elements of FM such that q1 ' +1, q2 ' +1 and p1
q1 , p2
q2 doesn’t contain any standard rational number (in this case p1
q1 = p2
q2). Then there exist a …nite sequence of irreducible rational numbers li
mi i=1;2;:::;g
such that:
p1 q1 = l1
m1 < l2
m2 < ::: < lg mg = p2
q2 where li+1
mi+1 li
mi = " for i = 1;2; :::; g 1. Besides for i = 1;2; :::; g we
have "mi = 0and mi = +1.
Proof. Let us consider the case where p2 q2
p1
q1 is not of " form; otherwise the lemma is proved. Let ti
i i=1;2;:::;r
be the elements of FM such that p1
q1 = t1
1
< t2
2
< ::: < tr
r
= p2 q2. Let i0 2 f1;2; :::; r 1g such that ti0+1
i0+1
ti0
i0
is not of " form, because if a such i0 does not exist the lemma is proved. From the properties of FM ( 1.2.1), i0+1 and i0 cannot be equal. Then there are two cases:
A) i
0+1 > i
0 : Let us take, in this case, g0 = +1 an integer such that g0
i0
= 0 ( the existence of g0 is assured by Robinson’s lemma). Let X =
E g0
" i0 and
H = ti0
i0
; Up; Up 1; :::; U0 wherep=E X i0+1
i0
and Ui = ti0+1+i:ti0
i0+1+i: i0 (i= 0;1; :::; p 1; p). Now we prove that : p is an unlimited integer, the product of the denominator of every element of H by" is an in…nitesimal and the distance between two successive elements of H is of the " form.
Indeed, we haveX =E g0
" i
0
= g0
" i
0
X where X 2[0;1[.
X i0+1
i0
= g0
" i
0 i0
X i0
i0+1 i0
= g0 " i
0 X " i
0 i0+1
" i0 i0 .
Since " i0 X = 0, " i0 i0+1 is a limited real number otherwise ti0+1
i0+1
ti0
i0
= 1
i0 i0+1
=" what contradicts the supposition. Theng0 " i
0 X " i
0 i0+1
is a positive unlimited real. On the other hand " i
0 i0 is limited; then
X i
0+1 i0
is a positive unlimited real, therefore p is also. The greatest denominator in H is i0+1+p i
0 where p= g0
" i0 i0
X i0
i0+1 i0
with 2[0;1[.
" i0+1+p i0 = " i0+1+ g0
" i0 i0
X i0
i0+1 i0
i0
= " i
0+1+ g0
i0
" X " i
0+1 " i
0 = 0.
Hence the product of the denominator of every element of H by " is an in…nitesimal. It remains to prove that the distance between two elements of H is of the " form; Indeed: Let i2 f0;1; :::; p 1g, from (1:7) we have
Ui Ui+1 = 1
i0+1+i: i0 i0+1+ (i+ 1): i0 .
By hypothesis we have i0+1 > i0, then of properties of Farey’s series (1.2.1)) 2 i
0+1> i
0+1+ i
0 > M, then i0+1 > M 2 . Let di =" i
0+1+i: i
0 i0+1+ (i+ 1): i
0 . Seen that i0+1 2 > M
2
2
, di is unlimited, therefore Ui Ui+1 =" . To
…nish the proof, we have of (1:7):
Up ti0
i0
= 1
i0+1+p i0 i0.
Let dp = " i0+1+p: i0 i0, after the replacement by the value of p, we obtain
dp =" i0+1 i0 +g0 " X i0 " i0+1 i0 " i0 i0. Since " i
0 = 0, " i
0 i0 is limited, then dp is unlimited; hence Up ti0
i0
=" . Thus, we end what we perceived.
B) i0 > i0+1: Let us take, in this case, g1 = +1 an integer such that g1
i0+1
= 0 (the existence of g1 is assured by Robinson’s lemma). Let Xe =
E g1
" i0+1 and
He = V0; V1; :::; Vp0 1; Vp0; ti0+1
i0+1
where p0 = E Xe i0
i0+1
!
and Vj = ti0 +j:ti0+1
i0 +j: i0+1 ( j = 0;1; :::; p0 1; p0).
Since the symmetry of this case with the case A) we prove, as in the case of H, that p0 is an unlimited integer, the product of the denominator of every element ofHe by"is an in…nitesimal and the distance between two successive elements of He is of the " form.
Thus the elements of H (or of He ) form a subdivision of the inter- val ti0
i0
; ti0+1
i0+1
. For the other intervals ti
i
; ti+1
i+1 i2f1;2;:::;r 1g fi0g
which don’t have a length of " form we do the same construction as we did with ti0
i0
; ti0+1
i0+1
.
By regrouping rational numbers which subdivide intervals ti i
; ti+1 i+1
(i 2 f1;2; :::; r 1g) not having a length of the " form and the rationals which are borders of intervals having a length of the " form, we obtain the
…nite sequence li
mi i=1;2;:::;g
. The irreducibility of the elements of the se- quence li
mi i=1;2;:::;g
results from properties of Farey’s series.
Lemma 2.5. Let 2[0 ,1]be a real, if is not in the"-galaxie of a standard rational number then there exists two irreducible rational numbers h1
k1;h2 k2 of the interval [0;1] such that
2 h1
k1;h2
k2 ; k1 = +1; k2 = +1; "k1 ="k2 = 0and h2
k2 h1
k1 =" .
such thatp
" = 0and letFM be the Farey sequence of orderM =E p" . Let p1
q1, p2
q2 be two successive elements of FM such that 2 p1 q1;p2
q2 . Two cases are distinguished:
A) Nor p1
q1 nor p2
q2 is a standard rational : In this case by applying the lemma 2.4, we obtain two irreducible rationals li0
mi0 and li0+1
mi0+1 such that 2 li0
mi0, li0+1
mi0+1 , mi0 = +1, mi0+1 = +1, "mi0 = "mi0+1 = 0, li0+1 mi0+1 li0
mi0 =" . Hence the lemma is proved by taking li0
mi0 for h1
k1 and li0+1
mi0+1 for h2 k2. B) p1
q1 or p2
q2 is standard (cannot be both at the same time standard). Let us suppose that p1
q1 is standard (the other case, seen the symmetry, can be treated by the same way.). Then p1
q1 ="w where w = +1. Let us put
L=E 2= p1
q1 then "L= 0and p1 q1 + 1
L < . Let l
m be the reduced form of p1
q1 +1
L, then"m= 0because m Lq1 andq1 is a standard. m > M because l
m is not an element of FM. Therefore "m2 is an unlimited because
"m2 > "M2 and "M2 is an unlimited. This means thatm is of theE
0
p"
!
form where 0 is a positive unlimited real verifying p
" 0 = 0. Now if we consider Fm, then 2 p01
q01;p02
q20 where p01
q10 and p02
q20 are two successive non standard elements ofFm. Thus the caseB)comes back itself to the caseA), therefore the proposition is also proved for this case.
Remark. We easily see that this proof is also a proof for the theorem 1.5.
Let be a positive unlimited real such that": '0, then Lemma 2.6. There exists a …nite set
S =fl1; l2; :::; lng [0;1] (2.5) containing all standard elements of [0;1] such that jli+1 lij " for i 2 f1;2; :::; n 1g.
Proof. Let B(S; z) be the internal formula: "S [0;1] is …nite, z 2 S &
8(x1 , x2) 2 S S (jx1 x2j " )". Let Z [0;1] be a standard …nite part. Then there exists a …nite partS [0;1]such that every elementz ofZ belongs to S and 8(x1 , x2)2S S (jx1 x2j " ), i.e. we have B(S; z).
Indeed it su¢ ces to take S = Z. Therefore, the principle of idealization (I) asserts the existence of a …nite part S [0;1] such that any standard element of [0;1]belongs to S and 8(x1 , x2) 2S S (jx1 x2j " ). Put S =fl1; l2; :::; lng, wherejli+1 lij " fori2 f1;2; :::; n 1gand any stan- dard element of [0;1]belongs to S.
Corollary 2.7. For every element li of S (S is the set that has been con- structed in the lemma 2.6 ) we have only one of the two cases:
1) li is a standard rational number.
2) li is outside of " galaxies of all standard rational number.
Proof. Letli 2S, then
1) li can be a standard rational because S contains all standard elements of [0;1].
2)li is not a standard rational thenli is not in the" galaxy of any standard rational. Indeed, suppose that li = p
q +"$ ($ 6= 0), where p
q is standard.
Then li and p
q are elements of S with li p
q =j"$j< " which contradicts lemma 2.6 .
Lemma 2.8. For every standard integer n 1. The real numbers xi of all system fx1; x2; :::; xng S (S is the set that has been constructed in the lemma 2.6.) are approximated by rational numbers Pi
Q i=1;2;:::;n
to" near with "Q = 0. that is to say:
8<
:
xi = Pi Q +"
"Q = 0
;i= 1;2; :::; n. (2.6) Proof. Consider the formula:
A(n) "8 fx1; x2; :::; xng S 9 Pi
Q i=1;2;:::;n
such that:
8<
:
xi = Pi Q +"
"Q = 0
;i= 1;2; :::; n ".
According to the corollary 2.7, a real x of S is a standard rational or is out- side of " galaxies of standard rationals. In addition, according to lemma 2.5, if x is not in the " galaxy of a rational standard, x is written in the form
8<
:
x = P
Q +"
"Q = 0
. Then in all cases x is written in the form 8<
:
x = P
Q +"
"Q = 0
. Consequently we haveA(1).
Suppose A(n), for a standard integer n, and prove A(n+ 1).
Let (x1; x2; :::; xn; xn+1) S. Since A is veri…ed for n we have ( xi = pi
q +" ; i= 1;2; :::; n
"q= 0
. (2.7)
If xn+1= h1
k1 is standard, then becausek1 is standard and of (2:7) we have 8>
>>
<
>>
>:
xi = pik1
qk1 +" = Pi
Q +" ; i= 1;2; :::; n xn+1 = h1q
k1q +":0 = Pn+1 Q +"
"Q="qk1 = 0
. (2.8)
Let us look at the case where xn+1 is not a rational standard. In this case the application of the theorem 1.5 to the real qxn+1 with the in…nitesimal"q implies:
(
qxn+1 = M
N + ("q)a ("q)N = 0
where a is limited. If a= 0, then from this and (2:7): 8>
>>
<
>>
>:
xi = piN
qN +" = Pi
Q +" ; i= 1;2; :::; n xn+1 = M
qN +"a = Pn+1
Q +"
"Q ="qN = 0
. (2.9)
Let us look at the case where a is appreciable. Suppose a >0, then 8>
>>
<
>>
>:
xi = N pi
N q +" ; i= 1;2; :::; n xn+1 = M
N q +"a
"N q = 0
. (2.10)
The reduced form of M
N q cannot be a rational standard. Otherwise, xn+1 and M
N q become two elements ofS such that the separating distance between them, is of the "a form. What, according to lemma 2.6, is not true for two elements of S; for the same reason xn+1 cannot be in the " galaxy of a standard rational. According to the lemma 2.5:
8<
:
xn+1= h1 k1
+" 1 = h2 k2
" 2
"k1 ="k2 = 0 ; k1 =k2 = +1
(2.11)
Where 1 0 and 2 0 are two in…nitesimal reals and h1 k1, h2
k2 are irre- ducibles. Let the element of S succeeding immediately xn+1 in S (xn+1 <
). Then by lemma 2.6 :
xn+1 ="! = 0,! . The real number xn+1+
2 is not in the " galaxy of a rational standard, otherwise, x and does not become two successive elements ofS. Hence,
according to the lemma 2.5 ( xn+1+
2 = s l " 4
"l = 0 ; l= +1; 4 0 and 4 = 0
. (2.12)
where s
l is irreducible. Let be an unlimited natural number such that p : = 0 and N = E p . Let us take N = max N ; k2; l . Then N = +1 and is of the E p form with is a positive unlimited real verifying p
: = 0. In the other hand h2
k2 and s
l are two elements of FN such that h2
k2
, s
l doesn’t contain any rational standard and k2 = +1 and l = +1. In this situation the lemma 2.4 is applicable and consequently there is a …nite sequence of irreducible rational numbers si
li 1 i e
such that h2
k2 = s1 l1 < s2
l2 < ::: < se le = s
l where e= +1and for i= 1;2; :::; e 1 we have :
si+1 li+1
si li
=" . Besides we have"li = 0,li = +1fori= 1;2; :::; e; se
le s1
l1 =" !
2 + 4 2 . In this paragraph we will associate to eachi 2 f1;2; :::; eg a vector Vi in Qn+1 such that then …rst components ofVi are in the "-galaxie of then …rst components of (x1; x2; :::; xn; xn+1), respectively. Whereas the (n+ 1) th component of Vi is equal to si
li. Indeed, for i = 1 apply lemma 2.2 to the system (l1x1; l1x2; :::; l1xn) with the in…nitesimal "l1:
8<
:
l1xi = Ti;1
t1 + ("l1)$ ; i= 1;2; :::; n l1t1 = 0
.
Hence 8<
:
xi = Ti;1 l1t1
+ "$ ; i= 1;2; :::; n
"l1t1 = 0
. Then
8>
<
>:
xi = Ti;1
l1t1 + "$ ; i= 1;2; :::; n
xn+1 = Tn+1;1
l1t1
1
(2.13)
whereTn+1;1 = s1t1, "l1t1 = 0 and 1 = " 2. Then we obtain the vector V1 = T1;1
l1t1 ; T2;1
l1t1 ; :::;Tn+1;1 l1t1
T
, where xn+1 = Tn+1;1 l1t1 = s1
l1.
Again the application of the lemma 2.2 to the system(l2x1; l2x2; :::; l2xn)with the in…nitesimal8 "l2, gives:
<
:
l2xi = Ti;2
t2 + ("l2)$ ; i= 1;2; :::; n l2t2 = 0
.
Hence 8<
:
xi = Ti;2
l2t2 + "$ ; i= 1;2; :::; n
"l2t2 = 0
. Then 8>
<
>:
xi = Ti;2
l2t2 + "$ ; i= 1;2; :::; n
xn+1 = Tn+1;2
l2t2 2
(2.14) whereTn+1;2 =s2t2 ,"l2t2 = 0and 2 =" 2+" with0< 1 < 2. Then we obtain the vector V2 = T1;2
l2t2 ; T2;2
l2t2 ; :::;Tn+1;2 l2t2
T
, where xn+1 = Tn+1;2 l2t2 = s2
l2.
Thus we construct the following vectors:
Vi = T1;i liti
; T2;i liti
; :::;Tn+1;i liti
T
; i= 1;2; :::; e (2.15) where for i= 1;2; :::; e: xn+1 = Tn+1;i
liti i = si
li i with "liti = 0.
Besides 0< " 2 = 1 < 2 < ::: < e= "!
2 +" 4 and for i= 1;2; :::; e 1:
i+1 i =" .
Lethbe the smallest integer such thathN q max
i (liti), then"hN q = 0.
On the other hand and according to Robinson’s lemma it exists an integer
W = +1 such that:
"W hN q = 0.
Put K =hN q. From (2:10):
8>
<
>:
xi = hN pi
hN q + " = Hi
K + " ; i= 1;2; :::; n
xn+1 = hM
hN q + "a = Hn+1
K + "a
(2.16)
where "K = 0, K max
i (liti).
LetW = min W, !
2 + 4 2 and Tn+1;i0
li0ti0 be the element of the sequence Tn+1;i
liti i=1;2;:::;e
which is the farthest from Tn+1;1
l1t1 verifying Tn+1;i0
li0ti0
Tn+1;1
l1t1 ="W
withW W. One notices thatW = +1because by constructionW W = .
Let R 1 be the integer such thatRli0ti0 K <(R+ 1)li0ti0. In this case Rli0ti0 and K are of the same order of magnitude i.e. : K
Rli0ti0 = where is a positive appreciable. Consider, the rationals of the following vector:
RT1;i0
Rli0ti0, RT2;i0
Rli0ti0, ..., RTn;i0
Rli0ti0, RTn+1;i0 Rli0ti0
T
. (2.17)
Where then …rst components of(x1; x2; :::; xn; xn+1)are in the" galaxies of the n …rst components of the (2.17), respectively. Whereas xn+1 is far from the last component of (2.17) by "W +" 2. We will search a positive integer j0 for which the rational RTn+1;i0 +j0Hn+1
Rli0ti0 +j0K becomes equal to Hn+1
K +"a+"
i.e. equal to xn+1+" . Indeed, put
j = RTn+1;i0 +jHn+1 Rli0ti0 +jK
Hn+1
K . (2.18)
Then j =
1 +j where is the distance between RTn+1;i0
Rli0ti0 and Hn+1 K which is equal to "W +" 2+"a.
Put 1 +j ="a. For this 1 +j =
"a. Hence
j = 1 "a
"a = 1 "W +" 2
"a
!
= W + 2
a = +1
.
Let us takej0 =E W + 2 a
!
, hencej0 = W + 2
a with 2[0;1[. Then
j0 =
1 +j0 . After the substitution by the value of and of j0:
j0 = a: "W +" 2+"a
a+W + 2 a
= "a W + 2+a
W + 2+a a
! .
Hence
j0 = "a:
W + 2+a
W + 2+a 1 a
W + 2+a
!
= "a 1
1 .
Since 1
1 = 1 + , then :
j0 ="a+" . (2.19)
On the other handj0 and W are of the same order of magnitude; indeed:
j0 W
= 1
W
W + 2 a
a
!
= 1 +
a .
Therefore j0 W
=A with A is appreciable, hence j0 =AW. Since W W one has: j0 =AW AW AW.
Lemma 2.9. The denominator of RTi;i0 +j0Hi
Rli0ti0 +j0K ( i= 1;2; :::; n+ 1) veri…es
"(Rli0ti0 +j0K) = 0 and for i= 1;2; :::; n; n+ 1 we have:
xi = RTi;i0 +j0Hi
Rli0ti0 +j0K +" (2.20)
Proof.
Rli0ti0 +j0K = K
+j0K
= K 1
+j0 . Hence Rli0ti0 +j0K K 1
+AW . From the fact that"W K = 0; A and are two appreciable numbers, we have "(Rli0ti0 +j0K) = 0. On the other hand for i = n+ 1 we have from (2.16) xn+1 = Hn+1
K +"a and from (2.18) and (2.19)
j0 = RTn+1;i0 +j0Hn+1 Rli0ti0 +j0K
Hn+1
K ="a+" . Hence RTn+1;i0 +j0Hn+1
Rli0ti0 +j0K " = Hn+1
K +"a =xn+1, this means that xn+1 = RTn+1;i0 +j0Hn+1
Rli0ti0 +j0K +" . For i= 1;2; :::; n we know from(2:17) that:
RTi;i0
Rli0ti0 xi ="$. (2.21)
Hence
RTi;i0 Rli0ti0
Hi
K = RTi;i0
Rli0ti0 xi +xi Hi K RTi;i0
Rli0ti0 xi + xi Hi
K ="$+" ="$ .
Therefore
RTi;i0 Rli0ti0
Hi
K = RTi;i0K HiRli0ti0
KRli0ti0 = $. (2.22)
Then we have: RTi;i0 +j0Hi Rli0ti0 +j0K
Hi
K = RTi;i0K HiRli0ti0 KRli0ti0 1 +j0: K
Rli0ti0
= $
1 +j0 . Since j0 = +1, then
RTi;i0 +j0Hi Rli0ti0 +j0K
Hi K ="
and seen that for i= 1;2; :::; n, the rational numbers Hi
K are, respectively, in the " halos of x1; x2; :::; xn then:
xi = RTi;i0 +j0Hi Rli0ti0 +j0K +" . So the lemma is proved.
Since "(Rli0ti0 +j0K) = 0, then if for i = 1;2; :::; n; n + 1 one takes RTi;i0 +j0Hi
Rli0ti0 +j0K for Pi
Q then 8<
:
xi = Pi
Q +" , i= 1;2; :::; n+ 1
"Q= 0
. (2.23)
In the case where a < 0 we take the element of S that precedes xn+1 i.e.
< xn+1 (S is ordered) and by doing, to a symmetry near, as we did for the case a >0.
From (2:8), (2:9) and (2:23) we have A(n+ 1). Hence, according to the ex- ternal recurrence principle, the lemma 2.8 is proved.
Let us return to the proof of theorem 2.1 De…ne forZ =fx1; x2; :::; xsg [0;1], the formula:
B(Z) = "9 Pi
Q i=1;2;:::;s
such that :8stm2N G Z, Pi
Q i=1;2;:::;s
, m
!
"
whereG Z, Pi
Q i=1;2;:::;s
, m
! 8
><
>: 1
" xi Pi Q
1
m ; = 1;2; :::; s j"Qj 1
m
is in- ternal.
Consider the set
L=fn2N :n jSj &8s 2 f1,:::,ng 8Z =fx1; x2; :::; xsg S :B(Z)g. (2.25) where S is the set that has been constructed in the lemma 2.6 .Then
L=
8>
<
>:
n 2N :n jSj &8s2 f1,:::,ng 8Z =fx1; x2; :::; xsg S, 9 Pi
Q i=1;2;:::;s
8stm2N G Z, Pi
Q i=1;2;:::;s
, m
! 9>
=
>;. According to lemma 2.8, L (N ) . If L is internal then, according to the Cauchy principle, it must contain (N ) strictly and therefore there is an integer ! = +1 and ! 2 L. If L is external then by the idealization principle (I) we can write L as follows:
L=
8>
<
>:
n 2N :n jSj &8s2 f1,:::,ng 8Z =fx1; x2; :::; xsg S, 8stf ini M 9 Pi
Q i=1;2;:::;s
8m2M G Z, Pi
Q i=1;2;:::;s
,m
! 9>
=
>;. whereM belongs to the set of …nite parts ofN . Therefore, Lis an halo ([4], [6]). Of the fact that(N ) L and no halo is a galaxy (Fehrele principle), then (N )
6
= L. Hence it exists an integer != +1and !2L.
Consequently in the two cases (L internal or external ) we …nds that it exists an integer ! = +1 and ! 2L, this signi…es that! jSj.
By lemma 2.3, there is a …nite part F [0;1] containing all standard elements of [0;1] such that jFj = !0 = +1 and !0 < !. Then F \S is a …nite part of S containing all standard elements of [0;1] with jF \Sj jFj = !0 < !. Put F \ S = fx1; x2; :::; xn0g. Then 9 Pi
Q i=1;2;:::;n0
such
that : 8<
:
xi Pi Q ="
"Q= 0 ; i= 1;2; :::; n0
. It follows that if x 2 R is a standard then x E(x) = Pi1
Q +" wherei1 2 f1;2; :::; n0gsince x E(x) is a standard of [0;1]. Hence
x = E(x) + Pi1
Q +" = Px Q +"
where "Q= 0. Thus the proof is complete.
3 Deduction of the classical equivalent of the main result
The theorem 2.1. can be written as follows
8" 8str (0< " r) =) 9q 8stx 8stt (kqxk< "qt &"q t)
where ", r 2 R +, q 2 N, x 2 R and t 2 R +. By using the idealization principle (I), the last formula is equivalent to
8" 8str (0< " r) =) 8st f iniX 9q 8(x, t)2X (kqx k< "qt& "q t) where X belongs to the set of …nite parts of R R +. This last formula is equivalent to
8st f iniX8"9strf(0< " r) =) 9q 8(x, t)2X (kqx k< "qt& "q t)g. Again, by using the idealization principle (I), the last formula is equivalent to
8st f iniX 9st f ini R 8" 9r2R f(0< " r) =) 9q 8(x, t)2X (kqx k< "qt& "q t)g.
where R belongs to the set of …nite parts of R +. By the transfer principle (T), this last formula is equivalent to
8f iniX 9f iniR 8" 9r2R f(0< " r) =) 9q 8(x, t)2X (kqxk< "qt &"q t)g. This last formula is exactly the main theorem announced in the abstract.
Indeed, if X = f(x1, t1), (x2, t2), ..., (xn, tn)g is a …nite part of R R +, then there exist a …nite part R of R + such that for all " > 0 there exists r 2R such that if0< " r then there exist rational numbers pi
q i=1;2;:::;n
such that:
8<
:
xi pi q "t
"q t
;= 1;2; :::; n.
Ref erences
[1]A. BOUDAOUD,Modélisation de phénomènes discrets et approximations diophantiennes in…nitésimales, Maghreb Mathematical Review Vol 1(1), June 1992.
[2]F.Diener, G.Reeb, Analyse Non Standard, Hermann éditeurs des sciences et des arts . 1989 .
[3] G.H. HARDY & E.M. WRIGHT, An introduction to the theory of num- bers, Clarendon Press, Oxford, 1979.
[4] F. Koudjeti, Elements of External Calculus with an application to math- ematical Finance, PhD thesis, Groninjen (The Netherlands), 1995.
[5]R. Lutz, M.Goze, Non Standard Analysis. A practical guide with applica- tions, Lecture note in Math. N0881, Springer Verlag (1981).
[6]E. Nelson, Internal set theory : A new approach to non standard analysis, bull. Amer. Math. Soc. 83(1977) 1165-1198.
[7] W.M. SCHMIDT,Diophantine approximation, Lectures Notes in Mathe- matics, N0785, Springer-Verlag, berlin, (1980).
[8] I. P. Van den Berg, Nonstandard Asymptotic Analysis, Volume 1249 of Lecture Notes in Mathematics. Springer Verlag, 1987.